Research On Individual Role Recognition Based On Information Interaction Network

21st century is the era of complexity.As a powerful tool for the study of complex systems,complex network has gained remarkable theoretic and practical progress and generated significant social influence after more than ten years of development.As a branch of the complex network research,social network study,with the widespread use of social media and the recent availability of large data sets of human interactions,has gradually drawn much attention.Social Network Analysis(SNA)has achieved positive practical effects in the field of transmission control of infectious diseases,crime organization disruption,domain expert recommendation,opinions leader mining in social media,rumor source detection and military organization structure identification.And these problems all can be abstracted as social networks in the section of the role analysis and identification,the studies of this problem thus has great practical significances.This study focuses on the study of individual’s role recognition based on the information interaction network,which mainly contains the following contributions:(1)Establishment of model framework for individual role recognition oninformation interaction networkRole decides the behavior.Based on this understanding,this study analyzed the individual’s role from the interaction between the members of the organization.By computing the semantic similarity between the interactive content and the specific topic,the weight of the interaction edge could be determined,and so as the further weighted information interactive network;with the iterative formula and clustering algorithms,individuals’ role correlations could be calculated and the specific role the organization could as well be identified.A general role analysis model framework was established,which was: 1)define the specific role and determine the specific topic;2)extract topics from interactive contents;3)calculate semantic correlations between interactive contents and specific topics;4)quantify edge weights;5)compute local role correlations using neighboring information;6)compute global role correlations with clustering algorithms;7)specific role recognition.(2)Quantification of edge weight of information interactive networkIndividuals in the organization were abstracted as nodes,and interactions between the individuals were abstracted as edges,and thus an information interaction network was constructed.Based on ART topic model,the topics from interactive contents were extracted.And semantic similarities between interactive contents and the specific topic were calculated by using the conditional probability obtained from topic models,and then the semantic similarities were used as edge weights.(3)The local role’s correlation calculation based on local informationBased on four hypotheses,the local information of the individuals in the network was used to construct a formula for calculating the local role’s correlation,and it was proved to be the qualified formula according to requirements from the hypotheses and it restrained a unique stable solution.And the stable solution is independent of the initial states,which indicated that it may be of great help in practical uses.(4)Clustering algorithm to calculate the global role’s correlationBased on the soft clustering,namely the fuzzy K means clustering(FCM),and the hard clustering,i.e.,the density and distance(DD)clustering,the global role of the individual was calculated.Clustering algorithm can make full use of the global information,and the fuzzy K meant clustering gave a probability value list for all the individuals;DD clustering is the individual role classification results(whether it is a specific role).Two methods complement with each other,DD algorithm can help determine the threshold of FCM results,and the results of FCM algorithm can quantify the role of the individuals in DD algorithm.Finally in this study,experiments were carried out on the Enron dataset and the ICM-C82 dataset.Results verified that the model could effectively identify the members of the Raptor project in Enron Company and the criminals involved in the fraud activities.